Soil spectral inference in the Western Highveld, South Africa: evaluating spiking as a tool for improved modeling accuracy
Global soil spectral libraries are a cornerstone of precision agriculture, yet their predictive accuracy is often limited in underrepresented regions like South Africa due to a lack of local data. This study evaluates a spiking strategy –where the global Open Soil Spectral Library (OSSL) is augmente...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IOP Publishing
2025-01-01
|
| Series: | Environmental Research Communications |
| Subjects: | |
| Online Access: | https://doi.org/10.1088/2515-7620/adf698 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1849390875305574400 |
|---|---|
| author | A Kock G M Van Zijl |
| author_facet | A Kock G M Van Zijl |
| author_sort | A Kock |
| collection | DOAJ |
| description | Global soil spectral libraries are a cornerstone of precision agriculture, yet their predictive accuracy is often limited in underrepresented regions like South Africa due to a lack of local data. This study evaluates a spiking strategy –where the global Open Soil Spectral Library (OSSL) is augmented with local data– to improve soil property prediction in South Africa’s Western Highveld. Using mid-infrared (MIR) spectroscopy, we created a local dataset and spiked the OSSL at varying levels. We then employed machine learning models to predict key agricultural soil properties (extractable Ca, K, Mg, Na, and P as Bray-1). Results showed that spiking dramatically improved prediction accuracy; a one-fold spiking level reduced the Root Mean Square Error (RMSE) for Ca from 1128.45 mg.kg ^−1 to 46.09 mg.kg ^−1 and for Mg from 206.92 mg.kg ^−1 to 45.12 mg.kg ^−1 . Despite this success, locally calibrated models remained superior, achieving an R ^2 of 0.84 for Ca compared to the best spiked model’s 0.51. We also found that excessive spiking yielded diminishing returns, with the prediction error for some properties increasing at higher spiking concentrations. This spiking approach offers a scalable method to enhance global spectral libraries for data-sparse regions, but our findings underscore that it complements, rather than replaces, the development of local calibrations for achieving the highest accuracy in sustainable land management. |
| format | Article |
| id | doaj-art-d06ac9aa0a2c4ea59984f9f0320f3ef5 |
| institution | Kabale University |
| issn | 2515-7620 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | IOP Publishing |
| record_format | Article |
| series | Environmental Research Communications |
| spelling | doaj-art-d06ac9aa0a2c4ea59984f9f0320f3ef52025-08-20T03:41:15ZengIOP PublishingEnvironmental Research Communications2515-76202025-01-017808500410.1088/2515-7620/adf698Soil spectral inference in the Western Highveld, South Africa: evaluating spiking as a tool for improved modeling accuracyA Kock0https://orcid.org/0000-0001-5349-9798G M Van Zijl1https://orcid.org/0000-0001-5003-1081Unit for Environmental Sciences and Management, North-West University , Potchefstroom, North West, South AfricaUnit for Environmental Sciences and Management, North-West University , Potchefstroom, North West, South AfricaGlobal soil spectral libraries are a cornerstone of precision agriculture, yet their predictive accuracy is often limited in underrepresented regions like South Africa due to a lack of local data. This study evaluates a spiking strategy –where the global Open Soil Spectral Library (OSSL) is augmented with local data– to improve soil property prediction in South Africa’s Western Highveld. Using mid-infrared (MIR) spectroscopy, we created a local dataset and spiked the OSSL at varying levels. We then employed machine learning models to predict key agricultural soil properties (extractable Ca, K, Mg, Na, and P as Bray-1). Results showed that spiking dramatically improved prediction accuracy; a one-fold spiking level reduced the Root Mean Square Error (RMSE) for Ca from 1128.45 mg.kg ^−1 to 46.09 mg.kg ^−1 and for Mg from 206.92 mg.kg ^−1 to 45.12 mg.kg ^−1 . Despite this success, locally calibrated models remained superior, achieving an R ^2 of 0.84 for Ca compared to the best spiked model’s 0.51. We also found that excessive spiking yielded diminishing returns, with the prediction error for some properties increasing at higher spiking concentrations. This spiking approach offers a scalable method to enhance global spectral libraries for data-sparse regions, but our findings underscore that it complements, rather than replaces, the development of local calibrations for achieving the highest accuracy in sustainable land management.https://doi.org/10.1088/2515-7620/adf698soil spectroscopylocal prediction modelsmachine learningcubistmid-infrareddata augmentation |
| spellingShingle | A Kock G M Van Zijl Soil spectral inference in the Western Highveld, South Africa: evaluating spiking as a tool for improved modeling accuracy Environmental Research Communications soil spectroscopy local prediction models machine learning cubist mid-infrared data augmentation |
| title | Soil spectral inference in the Western Highveld, South Africa: evaluating spiking as a tool for improved modeling accuracy |
| title_full | Soil spectral inference in the Western Highveld, South Africa: evaluating spiking as a tool for improved modeling accuracy |
| title_fullStr | Soil spectral inference in the Western Highveld, South Africa: evaluating spiking as a tool for improved modeling accuracy |
| title_full_unstemmed | Soil spectral inference in the Western Highveld, South Africa: evaluating spiking as a tool for improved modeling accuracy |
| title_short | Soil spectral inference in the Western Highveld, South Africa: evaluating spiking as a tool for improved modeling accuracy |
| title_sort | soil spectral inference in the western highveld south africa evaluating spiking as a tool for improved modeling accuracy |
| topic | soil spectroscopy local prediction models machine learning cubist mid-infrared data augmentation |
| url | https://doi.org/10.1088/2515-7620/adf698 |
| work_keys_str_mv | AT akock soilspectralinferenceinthewesternhighveldsouthafricaevaluatingspikingasatoolforimprovedmodelingaccuracy AT gmvanzijl soilspectralinferenceinthewesternhighveldsouthafricaevaluatingspikingasatoolforimprovedmodelingaccuracy |